One of the most provocative headlines from the recent UKJT Statement on Liability for AI Harms was the suggestion that solicitors may in future be held negligent for failing to use generative AI (genAI). With some firms enthusiastically embracing genAI and others regarding it as an overhyped white elephant, both face risks around genAI adoption. In this article, we examine the professional indemnity risks faced by solicitors arising from the use of text-generating genAI (ie Large Language Models) by complainants and law firms. The use of AI or machine learning in disclosure is well established and gives rise to different issues and, like the risks involved in using agentic AI (ie tools which have access to other computer applications and can carry out tasks outside their interfaces on the user's behalf), is outside the scope of this article.
Use of AI by complainants
Solicitors will have started noticing complaint correspondence, pleadings and other court documents that appear to have been prepared using genAI. The hallmarks of AI-generated content include unnecessary detail, grandiose terminology and obvious giveaways like "meta" text (eg "That's a great question! Here's a paste-ready complaint to your solicitor…") or links ending with "?utm_source=chatgpt.com", indicating that the user located the relevant source using ChatGPT. While it's often impossible to be certain that you are looking at an AI-generated document, the best way to prepare is to experiment yourself, which will also equip you for decisions about how to use genAI in your practice. Even this is not ethically straightforward because of its environmental impact.
There is no reason in principle why a LiP should not use genAI to help draft complaints and claims. The key challenges for law firms are twofold: first, the widespread availability of free genAI tools drastically reduces the time commitment from claimants, increasing the risk of complaints or claims. Second, genAI-drafted complaints may demand more time from the solicitor because of the need to investigate a legally coherent claim or an unnecessarily long complaint.
First-tier complaints (FTCs) arguably impose the greatest burden as firms are expected to investigate and respond to these themselves, and there are no formal rules governing the form or length of a complaint. Firms can make clear that it will take longer to respond to a lengthy and detailed complaint and, if they have their own secure genAI tools, deploy them in responding to the FTC. Formal letters of claim must be "detailed" under the Professional Negligence Pre-Action Protocol. However, more prescriptive rules apply once proceedings are issued with CPR rules 16.2 and 16.4 requiring that the claim form and particulars contain "a concise statement" of the nature of the claim and facts. This could form the basis of an application for strike out or an order that the claimant amend their particulars. If an issued claim is successfully defended, solicitors should recover their costs, but if fixed recoverable costs (FRC) apply, these may not reflect the time spent. While the firm could argue that the claimant's behaviour is unreasonable, this is a high bar under FRC ("conduct for which there is no reasonable explanation" under CPR 45.23). With senior judges such as Lord Briggs and Lady Chief Justice Baroness Carr appearing supportive of the use of genAI by LiPs, this could be difficult to prove.
Use of AI by solicitors
While there is pressure from clients, competitors and staff to use genAI, there are grounds for legitimate concerns about its safety. We know that genAI frequently gets it wrong, exposing firms to the risk of reputational damage and regulatory sanctions; case law suggests that it can endanger privilege; and 90% of businesses reported that genAI had no impact on employment or productivity.
Despite the wide availability of genAI tools since late 2022 and the obvious regulatory implications, the SRA's guidance was limited and largely reactive until this month, when the SRA released a warning notice on the misuse of AI. The warning notice highlights the rules most relevant to genAI and reveals that the SRA has received a number of self-reports from solicitors who have relied on AI tools that have generated inaccurate or misleading content, and has also been warned of potential breaches by senior members of the judiciary. The warning highlights that solicitors and regulated individuals who use AI during the course of their work remain accountable for their work, including any output produced by genAI and work carried out by others under their supervision. Solicitors should also familiarise themselves with the Law Society's Guide to Generative AI.
However, the warning notice and Law Society guide do not lay out prescriptive requirements, so solicitors must still decide for themselves whether the potential rewards outweigh the risks. As mentioned, the best way to judge how genAI can aid your practice is to experiment with it (first opting out of your inputs being used to train the model).
One of the first considerations is security and confidentiality. Due diligence should include exploring where data will be processed, in what circumstances it will be retained by the provider and for how long. While publicly available genAI chatbots such as ChatGPT or Gemini are available for free, the best solution is likely to be an enterprise (ie business) version of a secure system, ensuring that firm and client data stays within the firm's boundary and is not used to train external models or shared with other users.
The Upper Tribunal in Munir v Secretary of State for the Home Department [2026] UKUT 81 (IAC) treated inputs into publicly available genAI tools as equivalent to publication to all the world, meaning that confidentiality and privilege are lost. While some commentators regard this as wrong in law, it remains the only E&W case on point and the warning notice does not acknowledge this perspective. Firms' terms of business and privacy notices should warn clients of the risks of genAI use, and of the firm's proposed use of genAI.
Firms should also consider an AI policy. This is a valuable opportunity to set out what the firm regards as proper use of genAI, which tools are permitted and for which types of data. This helps address the so-called "shadow AI" risk (unauthorised or ungoverned use of genAI by employees), which tends to occur when employees are unsure what is permissible. The policy should address whether genAI use needs to be disclosed by supervisees. Oral discussion of drafts is now more important as this will better enable the supervisor to judge whether the supervisee has understood the issues. GenAI should not be relied on for legal research due to the risk of hallucinated citations; the regulatory implications of this risk were highlighted in the leading case of Ayinde v Haringey [2025] EWHC 1383 (Admin).
Outputs should be independently verified against trusted sources. The more important the output, the more carefully it should be verified. Despite the humanising language often applied to it, genAI does not understand the meaning of words. Instead, it has used machine learning to map language to a numeric system reflecting how closely words are related to other words. As it is, in essence, a highly sophisticated predictive-text machine, it will not always adequately reflect the levels of nuance and detailed textual analysis necessary for competent legal practice. Outputs should be saved to the matter file where used to prepare legal advice or documents.
Professional indemnity insurance issues
GenAI gives rise to new flavours of familiar claims, as well as wholly new exposures. Familiar claims such as giving incorrect advice, missing limitation deadlines, following an incorrect court procedure or data breaches may now feature AI. Failure to verify AI outputs may be treated with less sympathy than a human misunderstanding of the rules. Such claims will perhaps increase as younger generations are both more likely to use genAI and less likely to have the experience to spot errors.
In the coming years, there are likely to be cases where the reasonably competent solicitor would have used genAI, but the defendant did not. At present, these instances may be limited to use of AI in disclosure, but this is likely to change as genAI becomes widely adopted and ways of working change.
The risk of including hallucinated citations in court documents is essentially new and one of the most well-known, with a staggering 1,868 cases worldwide featuring hallucinated content. While knowledge of the risk makes it easier for practitioners to mitigate, it may make such errors more difficult to defend. Similarly, wholly genAI-driven processes underlying higher volume work may give rise to novel PI issues as the definition of private legal practice is redefined for the AI age – especially where there is no "human in the loop" or advice is given directly by genAI chatbots. We may also see disputes over the meaning of the cyber exclusion, which could be further refined by the SRA as the technology used by law firms evolves. There is also scope for aggregation issues to arise where repeated mistakes are made.
The best way to reduce the risk of coverage issues is to be candid with insurers about genAI use, both at renewal and on notification, remembering to keep insurers updated on genAI adoption during the policy period. Insurers will be interested in risk management measures, such as policies and staff training.
Copyright & permissions
.jpg&w=3840&q=60)